Best AI chatbots

Reviewed 14 October 2026 against vendor help centers. HumanRank has not run a hands-on benchmark of these assistants, so this guide ranks nothing by measured quality.

Short answer

ChatGPT is the broadest starting point: OpenAI documents web search on every ChatGPT plan, including Free. Add Claude when long documents drive the work, Gemini when your files already live in Google, and Perplexity when finding sources matters more than drafting. These picks come from documented scope and limits, not from tests we have run. Choose by trying the same three tasks in each assistant, and turn off model training before you upload anything sensitive.

Decision table

How the four compare on the same fields

ChatGPT

• Best for: a single assistant that covers drafting, analysis and quick research.
• Documented strength: OpenAI says web search is available on ChatGPT Free, Go, Plus, Pro, Business, Enterprise, and Edu. Each uploaded file can be up to 512MB, and text files are capped at 2 million tokens.
• Main limitation: the file upload FAQ caps uploads at 80 files every 3 hours, and Free users get 3 uploads per day. Outside ChatGPT Enterprise, ChatGPT reads only the text in uploaded documents and discards embedded images.
• Training control: turning off "Improve the model for everyone" stops new conversations being used for training. Rating a response with thumbs up or down can still make that whole conversation eligible, according to the data controls FAQ.

Claude

• Best for: reading, revising and reasoning over long documents and sets of files.
• Documented strength: Anthropic's help center allows 500MB per file and up to 20 files per chat. Free accounts can create up to five Projects.
• Main limitation: Claude analyzes both text and visual elements only in PDFs of 100 pages or fewer. Longer PDFs are handled as text. Free-plan usage limits reset every five hours.
• Training control: consumer chats are used for training only if you allow it, with two exceptions Anthropic lists: chats flagged for safety review and programs you explicitly join. The setting covers new chats, not old ones.

Gemini

• Best for: people whose documents, mail and calendars already sit in Google services.
• Documented strength: Google's plan page lists a 1 million token context window on AI Pro and AI Ultra. You can attach up to 10 files per prompt, with videos up to 2 GB.
• Main limitation: without an AI plan the context window is 32k tokens. Usage refreshes every 5 hours up to a weekly cap.
• Training control: with Keep Activity off, future chats are not used for training unless you send feedback. Chats that human reviewers have read are kept for up to three years, even after you delete your activity, according to Google's Gemini privacy help page.

Perplexity

• Best for: questions where you need to see where an answer came from.
• Documented strength: Perplexity says each answer includes numbered citations linking to the original sources.
• Main limitation: files are limited to 40 MB. The free tier gets only a "very limited" number of Pro Searches.
• Training control: AI Data Retention is enabled by default for Free, Pro and Max. Turning it off does not remove training data already collected.

For research-heavy work, also read best AI search tools.

Training settings differ more than features

Every assistant here lets an individual turn off model training, but the defaults and exceptions differ. Claude uses consumer chats for training only if you allow it in its privacy settings. ChatGPT and Perplexity train by default until you switch the setting off. Gemini keeps human-reviewed chats for up to three years whatever you delete. Change these settings before you upload client or student material, not after.

Business plans change the picture. OpenAI says it does not use ChatGPT Business, Enterprise or Edu workspace content for training by default. Check your own plan's terms rather than relying on consumer settings.

Accuracy is still your job

No chatbot answer should be repeated as fact until you have opened the source it relies on.

Independent research supports that rule. A study coordinated by the European Broadcasting Union and the BBC tested ChatGPT, Copilot, Gemini and Perplexity on news questions. It found that 45% of AI answers had at least one significant issue, from wrong source attribution to factual errors. That study covered news answers in 2025, and assistants change often, so treat it as a warning about the category rather than a current score for any one product.

How to choose in one afternoon

1. Pick three tasks you repeat: a factual question with a known answer, a document you need summarized with its caveats intact, and a draft you need revised.
2. Run them in two or three assistants, starting with your own account settings, including training controls.
3. Score what you would have to fix: wrong facts, dropped exceptions and invented sources count more than tone.
4. Check the limits you will hit. Upload caps, context size and five-hour usage windows matter more on a deadline than on a demo.

Our ranking method explains how HumanRank will score these tasks once its own tests are published.

Related categories

More categories in the HumanRank category index: best AI writing tools, best AI coding tools and best AI search tools.

Method and limits

• How we rank and how we are funded: our ranking method explains what a future HumanRank test will measure, and our disclosure explains which links may earn money.
• Sources: we read each vendor's own help center and plan pages on 14 October 2026, plus two independent studies. Every figure on this page traces to one of those pages.
• Not tested: HumanRank has not yet run controlled tests of answer quality, speed or reliability. The "HumanRank score" column will stay "Not tested" until a published test exists.
• Prices: we left them out. Plan pages change often, and some did not state the billing period and currency clearly.
• What can change: limits, plan names and default settings. We re-check this page when vendors update their documentation.

Frequently asked questions

Which AI chatbot is best overall? There is no measured winner yet. ChatGPT covers the widest range of tasks on paper, but the right choice depends on your files, your tools and your privacy needs.

Which chatbot handles the largest files? ChatGPT documents the largest per-file limit at 512MB. Claude allows 500MB per file and up to 20 files per chat. Perplexity limits files to 40 MB.

Can I stop my chats being used for training? Yes, in all four assistants, through each one's settings. Defaults differ: ChatGPT and Perplexity train by default, while Claude uses consumer chats only if you allow it.

Does the free version of Gemini get the full context window? No. Without an AI plan the context window is 32k tokens. AI Plus raises it to 128k, and AI Pro and AI Ultra to 1 million.

Which chatbot cites its sources? Perplexity attaches numbered citations to every answer. Claude includes citations when web search is used. ChatGPT search shows sources too, and OpenAI warns that they can be incomplete or wrong.

Are AI chatbot answers reliable for news? Not without checking. The EBU and BBC study found significant issues in 45% of news answers across four assistants.

Is Claude free? Claude has a free plan with usage limits that reset every five hours. Paid plans are Pro, Max, Team and Enterprise.

Do business plans train on my data? OpenAI says business workspace content is not used for training by default. Check your contract, because terms vary by plan and vendor.